Database Reference
In-Depth Information
of case studies for the enhancement of existing DBMS with appropriate techniques to store multimedia
data and algorithms to efficiently execute similarity queries over them. Finally, we conclude the chapter
and give future research directions on multimedia retrieval support in DBMS.
QUERYING BY SIMILARITY
The Query Representation Problem
One of the reasons why the DBMS are widely adopted is related to its capacity to support the storage
and retrieval of large data volumes, enforcing concurrent and safety data access among several applica-
tions. They are also characterized by supporting efficient languages that allow both, the definition of
data structures (Data Definition Language - DDL) and the specification of queries and data updates
(Data Manipulation Language - DML), in such a way that the user has to worry in state what data is
wanted but not how to obtain them.
The Structured Query Language (SQL), the
de facto
standard language of DBMS, provides a plentiful
set of operators to represent queries involving “simple” data, such as numbers and short texts. However,
the modalities of querying operations based on multimedia content bring many more possibilities than
those feasible for simple data. If on the one hand searching over sets of numeric attributes usually allows
employing total order relationships or well-defined mathematical functions over them, on the other hand
searching over sets of multimedia data can involve non-trivial abstractions inferred from their content.
For example, the following queries are relevant for multimedia applications:
•
Q1
: return the images that have a mass of abnormal size in a radiological imaging exam;
•
Q2
: return the images that show a sunset over the sea;
•
Q3
: return the videos that contain parts of other copyrighted videos;
•
Q4
: return the videos of children playing with pets;
•
Q5
: return the songs that have a given drum solo.
Each of these queries requires a particular processing over the data content to be answered. The
point is: how to represent the query to be answered by a general-purpose retrieval system? By general-
purpose we mean a system that provides sets of primitives that can be reused and specialized by distinct
applications, and that share a common group of retrieval requirements. In the following subsections two
approaches to answer these problems are discussed.
The Content-Based Operators Approach
The strategy that aims at defining a specific function or operator to every abstraction identified on the
data content we call as the
Content-based Operators approach
. For example, to represent the query Q1,
this approach requires having a function that identifies a specific pattern in the images (a mass) and
another that calculates the area. In a similar way, Q2 demands functions to identify the desired elements
(sun and sea) in the scene and operators based on the spatial relationship among them (e.g. north, south,
east and west).
